Would this work for octane?

[jamestmather]

Hi guys,

I'm considering getting this:

http://www.sabrepc.com/p-2390-appro-app ... ebone.aspx


It boils down to 4x2090 teslas in rack mount system with CPUs on board. I need apparently to put some motherboard ram in it but after that I figure I can work on my laptop - fire the scene over the network and, using VNC set the beastie rendering while I go to bed. I realise I'd need to buy another octane and max plugin but so far it seems the fastest and safest bet. Cubix seem to be cagey about whether or not my laptop will work. From the tone of the cubix emails, I get the feeling that , if not compatible, then returning for a refund on that basis will be tricky.

I'm keen to go tesla as they are used to being run like a racehorse without fear of burnout plus the six gigs limit is nice. I also like that I can render while still working on my laptop. (plus 1300 gflops per card is nice). Are the 2090s good with Octane?

Any thoughts, wisdom or observations are greatly appreciated.

Thanks,

J.

[t_3]

should be good, becuase it is in fact just a slower gtx 580, at least in terms of sp precision - to bad, that octane doesn't need the fp performance of that card (4 times a gtx 580 ).

hopefully you have a soundproof room somewhere in the basement, because 4 of these crammed into a 1u case means that the fans will simulate a f15 take-off while octane is simulating more pleasant things

[jamestmather]

Thanks for the reply. Yes to the noise - although previous render farms have actually heated the house - I don't see why this should be any different.

Is it slower than the 580? I thought it's single point precision was 1664 Gflops - where as the 580 is 1581.1 - am I missing something?

J.

[gabrielefx]

should be good, becuase it is in fact just a slower gtx 580, at least in terms of sp precision - to bad, that octane doesn't need the fp performance of that card (4 times a gtx 580 ).

hopefully you have a soundproof room somewhere in the basement, because 4 of these crammed into a 1u case means that the fans will simulate a f15 take-off while octane is simulating more pleasant things

it's a lot expensive and noisy because uses fanless Teslas
Impossible to use this unit at home because needs a room at 18 celsius degrees or lower.

wait for the new Kepler and build yourself the box.

[t_3]

Thanks for the reply. Yes to the noise - although previous render farms have actually heated the house - I don't see why this should be any different.

Is it slower than the 580? I thought it's single point precision was 1664 Gflops - where as the 580 is 1581.1 - am I missing something?

J.

well, the simple math for all fermi tesla cards was (until now) sp = 2x dp performance; also tesla cards are usually lower clocked than their consumer counterparts, and with sp performance being equal across the whole fermi range (consumer and pro), the 1331 gflops sound correct to me...

wait for the new Kepler and build yourself the box.

this could be a long wait. high end kepler cards are scheduled for q3 imo...

[pixelrush]

I think I gave an opinion before but...
You are going to pay 8000 dollars more for 4xTesla m2090/6gb running hot and with vacuum cleaner noise than 4xGTX580/3gb under stressed using watercooling and near silent. The performance will be much the same.
With the difference you could buy another identical renderstation with Octane licenses and network render at some later time when/if Refractive implement that. So there is the possibility of twice as fast for the same money that way if maximum performance is very important.
If you didnt mind the heat and noise of aircooled and wanted to upgrade the cards later to keplers for say another 25% gain or perhaps waited for Maxwell with 55% more performance buy regular GTX580/3gb and even if you sold off the 'redundant' cards cheaply or just threw them out or one or two died early it would still be significantly cheaper than the Tesla route.
Unless you have a proven need for a full 6gb of vram I think you are crazy to buy Tesla, my opinion only.
Besides which each generation of nvidia cards offers more cuda functionality than the last.
It may well be that Refractive take advantage of newer cards to add some cool new interactive stuff you would like or perhaps they support 128 textures instead of 64.
You might come to regret paying so much for cards that don't allow these new capabilities and now are off the pace.
Graphic card development moves quickly. It might be better spend less each time and upgrade fairly often.
Perhaps 6gb will be the new 3gb as 512mb was once the new 256mb who knows.
Maybe those ideas help you make your choice.

[jamestmather]

Thanks Pixelrush - that helped a lot. I think I'll go for the 580s with watercooling as you suggested before (it's just my scenes seem to be stacking up a lot in terms of memory = up to 2gb before I knew what was going on and got a panic attack about memory. I have a skulltrail with 8 cores which I could add some 580's to pretty fast). Thanks for the feedback.

[pixelrush]

Octane is bound to vram capacity, don't get me wrong, its just 6gb comes at a premium price.
Everyone's work need is different however 3gb ought to be a fairly useful space.
Older Quadros FX5800 had 4gb but they were also expensive and only 240 cores so they arent a good idea now.
What sort of work do you do that is using up the 2gb of space easily and concerning you?
Is that attributed to many polys, large textures or image resolution? combination of?
Are you having to significantly compromise your scene/render to fit?
I think many people pay attention to their space management with Octane regardless of the card.
If you are a serious texture user then the number and size does require careful packing ATM its just a fact of gpu-cuda life. Perhaps if textures are the problem then the limited number wouldn't help even if you had the extra space? Maybe you could combine unwraps or something...
Anyway hope you managed to make a choice you will be content with until the next great thing comes along

[jamestmather]

Yeah. Going to go with 4xhydro coppers in the skulltrail as you suggested and stick to manageable texture sizes.

I tend to render at 2k for cinema size renderings . I've set up an industrial environment animation here which doesn't seem complex but escalated to nearly 2gb - lots of texture instances which I assume (or rather hope) that 3ds max octane instances rather than copies. I assume instanced geometry is simply copied by octane (must do a test).

Thanks for the sage advice Pixelrush. Appreciate it.

[pixelrush]

Geometry instances support is coming to Octane quite soon.
If that's regularly a significant part of your scenes you will be able to reclaim some valuable space and 3gb might not be so bad after all.
Happy rendering